How can we encourage compliance with on-street parking regulations to make better use of the public space and smooth the traffic flow? Parking violations are a problem common to most major cities worldwide, which see their urban centers undesirably submerged and wish to curb this phenomenon. The evolution of cultures and the emergence of new technologies, coupled with potential modifications in legislative frameworks, suggest a new management for parking. How can artificial intelligence leverage an optimize enforcement of on-street parking to work on the change in behaviors rather than repression? One of the answers with ParkPredict Control: a tool that uses data to improve on-street parking enforcement!
Context and issues
Organizational differences in North American countries
To start with the United States, parking enforcement can be carried out by the municipal police as well as private companies. Agents are in general efficient because they are paid according to a quota of controls to meet. Some cities also apply adaptive parking fares, like in San Francisco where rates are adjusted per block and time band, or in Seattle where fees are based on the demand per price zone. Such rules helped improve traffic and reduce the time wasted finding a parking space. However, they require smart communicating tools to efficiently control payment violations. A failure case occurs in Seattle, where a local phenomenon known as car ranching emerged, with several neighborhoods occupied by abandoned cars that stay for weeks or months.
Further north, in Canada, parking control functions are often performed by public enforcement officers, like in Montreal, Toronto or Vancouver. Some cities such as Calgary and Victoria rather delegate parking enforcement to private organizations for efficiency issues. Calgary further calculates responsive parking prices based on the difference between the observed demand and occupancy targets. This again illustrates the importance of technology for optimized parking management and enforcement depending on the local context.
Lastly, in Mexico, the national police is the entity that issues parking fines. However, parking enforcement is haphazard and rather targeted at obstructive parking. Moreover, corruption is still present in rural areas where agents ask for cash to close their eyes. Nevertheless, Mexico City has been a test site to experiment with new measures for the last decade. On-street parking in the city used to be a mix of regulated areas with parking meters, neighborhoods plagued by rogue valet services in front of shops, bars or restaurants, and sectors controlled by so-called franeleros who extorted drivers for a fee to park with the implicit threat that the car would be damaged if some tips were not paid. This resulted in a chaotic parking and street environment, with a significant amount of double parking and vehicles parked on sidewalks, corners or private doorways. The ecoParq program has drastically improved parking management in pilot areas by setting up an efficient pricing scheme, data collection infrastructures, and a structured enforcement. This is currently being extended to more and more areas in the city.
Review of reforms in European countries
Recently, the entry into force of the reform of on-street paid parking in France this year implied the decriminalization of enforcement and the decentralization of all associated functions. Once reserved for public agents such as the municipal police, ticketing can now also be carried out by certified third parties like private companies. The former criminal fine that was uniform throughout the national territory is also replaced by a fee for occupying the public domain, whose price ranges and enforcement procedures are fixed by each local authority. The reform has rapidly helped cities increase their direct payment rate, passing for example from 50 to 80% for Bordeaux and from 10 to 30% for Paris, even if some progress is still expected in the coming years.
Several European countries, such as Belgium, the Netherlands, the United Kingdom, Portugal, Spain, and Norway, had already taken the initiative and established similar decriminalization frameworks several years ago. These measures have generally been beneficial to local authorities if only in terms of compliance with parking rules and improved spontaneous payment. For example, the direct payment rate increased from 35 to 90% in the United Kingdom and from 30 to 90% in Spain following the reforms.
Another observed consequence is the reduction of road congestion by limiting the presence of overstaying vehicles, thus promoting access to city centers and reducing congestion problems. A positive side effect is, therefore, the diminution in greenhouse gas emissions and various pollutants, leading to a reduction in pollution and better air quality in cities. Finally, the financial gains of optimized enforcement can be reinvested in more sustainable and public mobility infrastructures such as public transport and alternative modes (self-service bikes and cars, car sharing, electric vehicles), and in the development of park and ride facilities.
A common insight from the various experiences of parking management in the world is that the organization of enforcement is a major lever of efficiency. In this context, making parking violations an administrative matter is crucial to achieving success. Indeed, it allows the adaptation of all enforcement issues according to local constraints that a centralized system would inappropriately neglect otherwise.
Another recurrent lesson to optimize enforcement functions is that the choice of technologies used is of utmost importance. In particular, enforcement can be facilitated with communicating systems for dematerialized payment and control. For instance, pay-by-plate mechanisms, both for parking meters and mobile payment applications, coupled with automatic number-plate recognition (ANPR) technologies for enforcement, are ideal to leverage efficient operations.
ParkPredict Control Technology
Problem, value proposition and alternatives
ParkPredict Control is a decision support system to optimize parking control. A major problem for control teams is to know where and when to patrol in order to carry out effective rounds.
Existing alternatives are based on traditional statistics in spreadsheets or other data visualization tools. For example, by plotting the temporal evolution of the non-payment rate or the map of positive and negative controls in the city, we obtain a graphical description of the past status of enforcement. These purely factual observations, however, are of limited value in predicting future rounds. A slightly more advanced diagnostic is to calculate temporal averages per day, as well as spatial averages per area in the city, in order to highlight the periods and sectors where the natural payment rate is lowest. However, this approach still focuses on a simple historical analysis of the data taken out of context and out of the dynamic environment of the city. Moreover, the amount of data required is prohibitive to work finely in logistics planning per street and per hour. Finally, these methods are often carried out by hand, which makes the work tedious, unreplicable, subject to miscalculation and misinterpretation.
To overcome these constraints, Qucit develops innovative technologies in artificial intelligence (AI). In particular, ParkPredict Control uses machine learning algorithms to predict where and when the number of violations is highest, and thus prescribe the areas to be visited by enforcement teams during the day. The machine is therefrom able to determine the ideal route to optimize parking enforcement and thus becomes a real recommendation engine. A major advantage is the ability to accurately forecast and extrapolate the number of violations from a reasonable amount of historical observations. Moreover, the machine learning algorithms allow the respective impacts of temporal and spatial variables to be decoupled, for recommendations by sector and period that are reliable, operable and operational on the field.
Solution, contextual data and machine learning
In more detail, our technological solution consists in enriching the control data (the positions and times of the controls performed and verbalizations issued), by adding a large number of contextual data flows. This contextual data digitalizes the environment to describe the city as realistically as possible according to the elements impacting usages and behaviors (for example, the number, location and type of parking spaces, ticketing data by parking meter or mobile payment, the position and occupancy of off-street car parks, points of interest such as restaurants, schools or shops, weather, calendar data such as time, day of week, bank and school holidays).
Once contextual modeling implemented, the machine learning calculation engines developed at Qucit work in two steps: calibration and prediction. During calibration, the models first train by learning the relationships between behaviors and context on all the data collected in the history. For example, the model assimilates that there are more violations on Fridays from 6 pm in a particular part of the city because many bars and restaurants are present, or on weekdays between 8 am and 9 am, excluding bank and school holidays, in other areas because there are schools. Once calibrated, models are able to provide predictions on the number of future violations for a given location, since they have already observed similar contexts in the past.
Each new control also generates data that is injected into our predictive models to verify and improve their reliability. Thus, the models are recalibrated periodically, adding the data collected since the last calibration to the existing data. Predictive models can thus improve their performance and adjust to changes in behavior. Data related to urban infrastructure or socio-economic variables are also updated regularly to adapt to structural changes in the city.
The recommendations can then made available to team managers in a dashboard for operations supervision and to agents on foot or in vehicles equipped with automatic number-plate recognition via a mobile application. Violations against paid, reserved, obstructive, illegal and dangerous parking may be included, as well as other behaviors such as contestation.
ParkPredict Control is thus positioned as an all-in-one tool for effective on-street parking enforcement powered by artificial intelligence. It is becoming easier to fight against the issues caused by non-payment and especially against congestion in urban centers.
- A comprehensive toolkit and study of on-street parking management in the world (English):
- A report on reforms and on-street parking enforcement in Europe with several case studies (French):
- Some presentation slides on our work on artificial intelligence for parking enforcement (English):